Hermes vs Copilot is not a fight over one seat: GitHub Copilot works on code inside GitHub, while Hermes Agent runs the always-on jobs around that code from a machine you control.
So the real job is wiring them together so they never step on each other.
Below is the install order, the settings I'd change first, the gotchas buried in each set of docs, and two small-business builds you can copy.
Neither vendor publishes a comparison with the other, so every split on this page is my own reading of the Hermes Agent docs and GitHub's Copilot docs.
Want my team to map the split for your own repos? Get your Backlink Gameplan, or try Polar Rank free for 3 days.
The setup view: Hermes and Copilot in one table
| Setting | Hermes Agent | GitHub Copilot |
|---|---|---|
| Made by | Nous Research, MIT license | GitHub |
| Runs on | Your laptop or server, with local, Docker, SSH, Modal, Daytona or Vercel Sandbox backends | Your IDE and terminal, a GitHub-hosted cloud sandbox, or a GitHub Actions runner |
| Install | One-line script on Linux, macOS and WSL2; PowerShell on Windows | npm, WinGet, Homebrew or an install script |
| You talk to it in | CLI, TUI, Desktop, Webapp, Telegram, Slack, WhatsApp, email and more | IDE chat, Copilot CLI, the Copilot app, github.com, GitHub Mobile, Slack, Teams |
| Timed work | Cron jobs that deliver to a chat, a file or another platform | Automations on a schedule, on repo events or through the API |
| Rules file | AGENTS.md in the working folder, plus SOUL.md | .github/copilot-instructions.md, plus AGENTS.md, CLAUDE.md or GEMINI.md |
| Output | A message, a file or a report | A branch, a pull request or a review |
Read the last row twice.
It tells you which tool owns which job.
What each tool is built to do
GitHub pitches Copilot as an assistant for the whole path from typing code to shipping it.
It completes lines as you type, explains a repo on request, reviews diffs and takes on assigned tasks.
Its natural loop is GitHub's own: an issue turns into agent work, the work returns as a pull request, and a person merges it.
Hermes Agent is an open-source agent from Nous Research that lives on hardware you choose and remembers you between sessions.
One gateway process connects it to your chat apps, runs its cron jobs and keeps every session in one place.
It writes code when asked, but code is one job among many, not the center of the product.
Step 1: Install Copilot CLI and give the repo its rules
- Confirm your Copilot plan; GitHub's plans page says every plan ships with Copilot CLI and the Copilot app.
- Install with
npm install -g @github/copiloton Node.js 22 or later, or use WinGet on Windows or Homebrew on macOS and Linux, as the install guide lists. - Open the repo folder and type
copilotto start an interactive session. - Press Shift+Tab to flip into plan mode before any task that touches more than one file.
- Create
.github/copilot-instructions.mdwith your stack, your test command and the folders it must leave alone. - Put rules both agents should follow in
AGENTS.mdat the repo root.
Copilot treats AGENTS.md as agent instructions, though GitHub notes not every Copilot feature reads that file yet.
Hermes loads the top-level AGENTS.md from its working folder when a session starts, so one file can brief both.
Step 2: Hand your first issue to the cloud agent
The Copilot cloud agent (formerly the Copilot coding agent) is the piece that keeps working after you close the laptop.
- Open the issue on GitHub and click Assignees.
- Pick Copilot from the list.
- Fill the optional prompt with constraints: files to avoid, the test that must pass, the style to match.
- Check the target repo, the base branch, any custom agent and the model in the dialog.
- Wait for the pull request; Copilot adds you as a reviewer when it finishes.
The cloud agent guide marks issue assignment as a public preview.
Prefer the terminal? The GitHub CLI page documents gh agent-task create to start a session and gh agent-task list to see your recent ones, on GitHub CLI v2.80.0 or later.
That list command matters later, because Hermes can run it for you on a timer.
🔥 Want the Copilot-plus-Hermes setup walked through on video? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members running Hermes next to their coding tools. → Join the AI Profit Boardroom
Step 3: Install Hermes and connect one chat app
- Run
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bashon Linux, macOS or WSL2, or the PowerShell installer on native Windows. - Run
hermes modeland pick a provider such as Nous Portal or OpenRouter. - Run
hermes gateway setupand choose Telegram or Slack. - Start it with
hermes gateway start, orhermes gateway installto keep it running as a user service. - Message the bot and confirm it answers.
Since v0.23.0, "The Hearth Release" of 5 November 2026, one gateway owns every session on a machine, so the CLI, Desktop, editors, bots and cron share a single history (release notes).
Desktop on that release needs macOS 13 or later, or 64-bit Windows.
Step 4: Let Hermes use your Copilot plan through the labs plugin
This is the only documented bridge between the two products.
In v0.23.0 Nous moved Copilot support out of Hermes core and into Nous Research's github-copilot labs plugin, which lives in the hermes-labs-plugins repo.
hermes plugins install github-copilot
hermes auth add copilot
The plugin gives you two providers.
| Provider | What it talks to | Sign-in |
|---|---|---|
copilot | The Copilot API with your GitHub account | hermes auth add copilot device code, or a token in the environment |
copilot-acp | Your installed Copilot CLI, driven over ACP with copilot --acp --stdio | The CLI's own copilot login |
Use copilot when you want Copilot's models behind a Hermes chat or cron job.
Use copilot-acp when you want Hermes to hand a turn to the Copilot CLI itself.
Profiles that already pointed at Copilot get the plugin installed on their next hermes update or gateway start, per the providers page.
Nous built it but labels it unsupported, and the catalog lists it in the community tier, so pin your Hermes version before you depend on it.
Step 5: Schedule Hermes jobs around the repo
Now give Hermes the work Copilot doesn't do: telling people what happened.
hermes cron create "0 8 * * 1-5" \
"In ~/sites/client-a run gh agent-task list. Summarize which Copilot sessions finished overnight and which pull requests wait for my review. Draft only: do not comment on GitHub." \
--name "Copilot morning digest" \
--deliver telegram
The cron docs let a job deliver to the chat it came from, a local file or any connected platform, and attach one or more skills.
You can pin a model per job, so the digest can run on a lighter model than your main chat.
Want Hermes to react the moment a pull request opens instead?
The webhook PR guide wires a GitHub webhook to a Hermes route that fetches the diff with the gh CLI and posts a comment.
That guide warns that pull request titles and descriptions can carry injected instructions, so run that gateway in Docker or on an SSH backend.
No public URL? The same guide points to a cron-based version that polls on a schedule instead.
📞 Not sure which repo jobs belong to Copilot and which to Hermes? Get your Backlink Gameplan and my team will go through your workflow with you, or try Polar Rank free for 3 days. Backlink Gameplan or try Polar Rank free for 3 days
The safety settings I'd change on day one
- Copilot MCP: once you add an MCP server, the cloud agent calls its tools without asking you first, so add only servers you'd trust with write access (MCP setup page).
- Copilot defaults: the GitHub and Playwright MCP servers are already switched on, so review them before adding more.
- Copilot firewall: internet access is limited by default, but the firewall only covers processes the agent starts through Bash, not MCP servers, and GitHub says not to treat it as a complete safeguard.
- Copilot CLI: skip
--allow-all-tools, which lets it run any shell command you could, and use/sandbox enableorcopilot --cloudfor risky work. - Hermes approvals: set
approvals.modeto smart or manual in config.yaml and leavecron_mode: deny, so a scheduled job stops at a dangerous command (security docs). - Hermes messages: approvals cover risky commands, not outgoing messages, so write "draft only" into any job that could reach a client.
Gotchas that eat an afternoon
- After you assign an issue, Copilot ignores new comments on that issue. Post changes on its pull request.
- Cloud or Actions is chosen when a session starts and can't be switched. Actions is the pick for setup steps, larger or self-hosted runners, or Windows.
copilot-setup-steps.ymlonly applies to Actions sessions and only triggers once it sits on your default branch.- Copilot connectors work in cloud-sandbox sessions only, not in Actions sessions such as those started by assigning an issue.
- Copilot works only with repositories hosted on GitHub, and it isn't available for GitHub Enterprise Server.
- Automations need a private or internal repo and one of Copilot Pro, Pro+, Max, Business or Enterprise.
- The labs plugin no longer reads tokens saved by the gh CLI or Copilot CLI, so run
hermes auth add copilotonce. Classicghp_tokens are rejected; a fine-grained token needs the Copilot Requests permission. - Hermes in ACP mode (
hermes acpinside VS Code, Zed or JetBrains) leaves out messaging delivery and cron management, so keep the schedule on the gateway.
Build 1: a three-person web agency
The agency keeps ten client sites in GitHub repos and a team chat in Slack.
- Each repo gets a
.github/copilot-instructions.mdwith the theme rules and the test command. - A client asks for a new pricing table; the account lead writes it up as an issue and assigns it to Copilot.
- Copilot works on its own branch and opens a pull request, then tags the developer as reviewer.
- The developer turns on Agent Merge's Watch and fix so Copilot handles failing workflows or merge conflicts (changelog).
- Hermes runs the morning digest from Step 5 for every repo and posts it to the team's Slack channel.
- On Fridays a second Hermes job drafts a plain-English change note per client, marked draft only, for the account lead to send.
Build 2: an accounting firm with one internal script
The firm has a small export script that turns bookkeeping data into a spreadsheet, kept in a private GitHub repo.
- The office manager opens Copilot CLI in that repo, switches to plan mode and asks for a new column.
- Copilot asks clarifying questions, writes a plan, then edits the script when told to go ahead.
- The change goes up as a pull request for the partner who owns the repo.
- Hermes runs a monthly cron job on the office server that reminds staff in Teams which client files are due.
- Client emails stay drafts: the job says draft only, and a person sends them.
A plumber or a salon with no code at all needs only the Hermes half of this page.
Where each one wins
- Copilot wins on anything that ends in a diff: completions in the editor, issues turned into pull requests, reviews, merge-blocker fixes and repo automations.
- Hermes wins on anything that ends in a person reading something: chat replies, scheduled summaries, reminders, memory of how you like things done, and any model you choose.
- Together, Copilot ships the change and Hermes tells the right people, and the labs plugin lets Hermes draw on your Copilot plan.
If Cursor is your editor rather than VS Code, my Hermes vs Cursor setup covers that pairing.
Verdict by job
- Pick Copilot if your code sits on GitHub and you want suggestions, reviews and pull requests from inside the tools your developers already open.
- Pick Hermes if most of your work is messages, schedules and research, or if you want to host the agent yourself and choose the model.
- Run both if you ship software and also run a business around it: Copilot on the repo, Hermes on the gateway.
🔥 Get the Hermes playbooks and the Agent OS in one place Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members swapping agent setups. → Join the AI Profit Boardroom
FAQ: Hermes vs Copilot
Can Hermes use my GitHub Copilot plan?
Yes, through the github-copilot labs plugin from Nous Research. Run hermes plugins install github-copilot, then hermes auth add copilot.
What is the difference between the copilot and copilot-acp providers?
copilot calls the Copilot API with your GitHub account. copilot-acp starts your installed Copilot CLI over ACP and uses that CLI's own login.
Why does Copilot ignore my new comments on an assigned issue?
GitHub's docs say Copilot only sees the issue as it was when assigned. Put follow-up changes as comments on the pull request it opens.
Can Hermes post to my clients automatically?
Hermes approvals cover risky commands, not outgoing messages. Write draft only into any job that could reach a client and send it yourself.
Do I need a public URL to connect Hermes to GitHub?
Only for the webhook route. The Hermes docs also describe a cron-based version that polls for pull requests on a schedule.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery".
- My Udemy courses have taught over 29,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Install Copilot CLI first, add the Hermes gateway second, and get your Backlink Gameplan if you want my team to check your hermes vs copilot setup, or try Polar Rank free for 3 days.











